Building population pharmacokineticpharmacodynamic models. I. Models for covariate effects

Abstract

One major task in clinical pharmacology is to determine the pharmacokinetic-pharmacodynamic (PK-PD) parameters of a drug in a patient population. NONMEM is a program commonly used to build population PK-PD models, that is, models that characterize the relationship between a patient's PK-PD parameters and other patient specific covariates such as the patient's (patho) physiological condition, concomitant drug therapy, etc. This paper extends a previously described approach to efficiently find the relationships between the PK-PD parameters and covariates. In a first step, individual estimates of the PK-PD parameters are obtained as empirical Bayes estimates, based on a prior NONMEN fit using no covariates. In a second step, the individual PK-PD parameter estimates are regressed on the covariates using a generalized additive model. In a third and final step, NONMEM is used to optimize and finalize the population model. Four real-data examples are used to demonstrate the effectiveness of the approach. The examples show that the generalized additive model for the individual parameter estimates is a good initial guess for the NONMEM population model. In all four examples, the approach successfully selects the most important covariates and their functional representation. The great advantage of this approach is speed. The time required to derive a population model is markedly reduced because the number of necessary NONMEM runs is reduced. Furthermore, the approach provides a nice graphical representation of the relationships between the PK-PD parameters and covariates.

What is the best size descriptor to use for pharmacokinetic studies in the obese?

“…The benefits of this approach lie in the predictive performance of the model for both interpolating PK responses that were not explicitly studied as well as predicting PK responses that arise from situations that exceed the scope of the original study (although care should be taken in this latter scenario). The interested reader is referred to Mandema [45], Wade [46] and Whalby [47,48] for reviews and considerations when building covariate models in population analysis.…”

“…Nevertheless, this method is widely used and is a valuable tool both in its own right and when used in conjunction with a fully population method (e.g. perhaps within a generalized additive modelling framework or using the Wald's approximation method) [44,45]. However, it is unfortunately common that results of such studies are published without reporting the coefficient estimates of the regression relationship, in these cases typically describing only the degree of association between covariate and parameter values as R 2 .…”

Semi-mechanistic population pharmacokinetic/pharmacodynamic model for neutropenia following therapy with the Plk-1 inhibitor BI 2536 and its application in clinical development

“…Once a model providing an adequate description of the data without the incorporation of covariates was selected, patient characteristics were explored for significance using the generalised additive model (GAM) approach [23] implemented in the software Xpose version 3 [24]. The covariates initially selected during the GAM analysis were further tested for significance in NONMEM using the forward inclusion and backward elimination approach.…”

Population PK and PK/PD modelling of microencapsulated octreotide acetate in healthy subjects

“…A stepwise generalized additive model based on the P i estimates from the basic population model as dependent variables was used to select the most important covariates and select the functional relationship between the covariate and the parameter [14]. The covariates assessed were age, weight, sex and baseline IGF-1 concentration.…”